Understanding variations in Robot Welding

Understanding variations in Robot Welding

Challenges and Pitfalls

TowardsProactiveControlofbothWeldQuality& Productivity

P.Hammersberg1,a,H.Olsson1,b1 ChalmersUniversityofTechnology,Dep.ofMaterialandManufacturingTech.,SE-41296Göteborg,Sweden2VolvoConstructionEquipmentAB,SE-67127Arvika,[email protected],[email protected]

usingLeanSixSigma

principle:Customerbasedoperationaldevelopment

Probleminanutshell:

Howtocontroltheweldingprocesssoproductdevelopmentdaretoreducemargins?

ExploringtheMAGweldingsystem

Remaining1. Mappingrelevantprocessfactors(p-diagram) à 24factors2. Qualitativefiltering(usingexperts) à 11

factors3. Quantitativescreeningexperiment(Saturated211-7=16runs) à 5factors

Surprisingscreeningresult:Allparametersrelatedtoweldingbeam(I,U,wirefeed,speed,etc.)definingtheWPS,wereoflessimportant fortheweldingdimensionsthantheweldinggeometry:steelplatesandweldinggunangles

ResearchQuestion:Ifthetraditionalbeamparameters(U,I,speed,threadfeed,etc.)

(normallyusedcontroleverything:productivity,metallurgyandoutgoinggeometryandquality)(withanormallyFIXEDweldinggeometryforplateandweldgunangles)

isoflessimportancethantheWELDINGANGLESforthefinalBEADSHAPEandQUALITY

Woulditthanbepossibleto:1. FirstsettheWPS-parameterstoreachnominalmetallurgyandproductivity

• WPS15*forthe5mmFilletWeldofinterest• includingadjustingcorrespondingU,I,travellingspeed,wirefeed,etc.

2. Shapetheweldbeadgeometryusingthesteelplateandweldgunangles?

ResearchAnswer:YES!

sofar….andthisprobablymeansalotofmoneyintheend

*15m/minwirefeed

Potentialresult

TodayssituationFactors

controlling Productivity Welddimension

Workorder

Weldbeam Tradeoff 2Weld

geometry Almost fixed:PBorPA 1

Tomorrowsdecoupled/uncoupled*situation:- towardsfunctionalindependence

Factorscontrolling Productivity Weld

dimension Workorder

Weldbeam

Factorset1 1

Weldgeometry

Factorset2 2

Axiomaticprocessdesign,HouseofQualitythinking(QFD)

*)Suh,NamPyo.(2001).“Axiomaticdesign,advancesandapplications”

WebToeRadius[mm]

Penetration[mm]

Beadthickness

WeldingGeometryControlFactors(x’s)

andBeadQualityResponses (y’s)

Gap[0mm,2mm]

Up(+)/Downslope(-)[-10°,20°]

GunangletoWebPlate[35°,55°]

Webplatetovertical[35°,55°]

Gunangleaboveunwelded seam[80°,120°]

Platethickness[10mm,40mm]

FlangeToeRadius[mm]

DeviationfromFlat[mm]

y =f(xi)

Process / Product

LCL

UCL

Target

USL

LSL

Ratherthanqualitybyinspectionsandpost-processing…Whencontrollimits(---)arewider

thanspecificationlimits(---)Increasedcostfromwaste&rework

andall-partinspection

Inspectionandpost-processingaddCOST

Process / Product

LCL

UCL

Target

USL

LSL

…qualityshouldbecontrolledonxfactorsupstream

Control factors

DrivingprincipleinQuality(robust)Engineering:Monitory,controlx!

Step1:IdentifyProductionSet-up

A. IdentifySet-up:• MAGwelding• Filletweld:

• 5mmthick(a)• 2mmpenetration(i)

• WPS15[wirefeedm/min]• Solidwire• Robotequipment• Etc.

B. Optimizeproductivity

Step2:Buildpredictionmodels:yi =f(xj)usingdesignofexperiments(DOE)

Controlfactors(x) Outputresponses(y)

Step2a:Selectexperimentaldesign

Run Thickness[mm]Up(+)/Down(-)slope[°]

Gunangelabove

unweldedseam[°]

WebPlateto

Horizontal[°]

GunangletoWebPlate[°] Gap[mm]

ToeradiusonFlange[mm]

ToeradiusonWeb[mm]

PenetrationdepthWeb

[mm]

PenetrationdepthFlange[mm] a-dim[mm]

Deviationfromflat[mm] Voltage[V] Current[A]

Wirefeed[m/min]

1 10 5 120 35 35 22 40 -10 120 55 35 13 40 -10 100 35 55 24 10 20 80 35 55 15 10 20 100 55 35 06 10 -10 120 45 55 07 40 20 80 45 35 28 10 -10 80 55 45 29 40 5 80 55 55 010 40 20 120 35 45 011 25 -10 80 35 35 012 25 20 120 55 55 213 25 5 100 45 45 1

DSDthickplate

InDSD:#runs=2x#factors+1**)PricewinningdesignofexperimentmethodologyfromBRADLEYJONESandCHRISTOPHERJ.NACHTSHEIMin2012

Forexample:a. Useaclassicalsequentialscreeningapproachor,b. DefinitiveScreeningDesign(DSD)estimatinguptosecond-ordereffects-

curvature – withoutcorrelations

Distributions

Input Control Factor DistributionsThickness [mm]

5

10

15

20

25

30

35

40

45

50

Up (+) / Down(-) slope [°]

-15

-10

-5

0

5

10

15

20

25

30

Gun angel above unwelded seam [°]

80

90

100

110

120

Web Plate to vertical [°]

35

40

45

50

55

Gun angle to Web Plate [°]

35

40

45

50

55

60

65

Gap [mm]

0

1

2

3

Ouput Respons DistributionsFlange Toe Radius [mm]

0

LSL

0,5

1

1,5

2

2,5

3

Web Toe Radius [mm]

0

LSL

1

2

3

4

5

Penetration [mm]

0

1

LSL

3

4

5

6

7

Bead Thickness [mm]

LSL

Target

5,5

6

6,5

Deviation from Flat [mm]

-2

-1

LSL

Target

USL

1

2

3

x’s:

Balanceddesign– alllevelsofthex’susedequallymanytimes

y’s:

Experimentalresult– rangeoftheresponsescovertheactualtargetrangeforallresponses.Potentialprocesswindowexist

Ouput Respons DistributionsFlange Toe Radius [mm]

0

LSL

0,5

1

1,5

2

2,5

3

Web Toe Radius [mm]

0

LSL

1

2

3

4

5

Penetration [mm]

0

1

LSL

3

4

5

6

7

Bead Thickness [mm]

LSL

Target

5,5

6

6,5

Deviation from Flat [mm]

-2

-1

LSL

Target

USL

1

2

3

Input Control Factor DistributionsThickness [mm]

5

10

15

20

25

30

35

40

45

50

Up (+) / Down(-) slope [°]

-15

-10

-5

0

5

10

15

20

25

30

Gun angel above unwelded seam [°]

80

90

100

110

120

Web Plate to vertical [°]

35

40

45

50

55

Gun angle to Web Plate [°]

35

40

45

50

55

60

65

Gap [mm]

0

1

2

3

Highersettingsonthesetwo

Darkgreenisconnected– largeradiusontheflangealsogenerallygavehighradiusontheweb.

x’s:

Balanceddesign– alllevelsofthex’susedequallymanytimes

y’s:

Experimentalresult– rangeoftheresponsescovertheactualtargetrangeforallresponses.Potentialprocesswindowexist

PredictiveModellingofResponseBehaviorone-by-oneExample:Penetration[mm](1of5outputs)

5,05533333333333

+ 0,879 * Gap [mm] - 1

+ -0,8098 *Gun angel above unwelded seam [°] - 100

20

+ -0,4364 *Up (+) / Down (-) slope [°] - 5

15

+ Gap [mm] - 1 * Gap [mm] - 1 * -1,1323333333333

+ Gap [mm] - 1 *Up (+) / Down (-) slope [°] - 5

15* 0,77825

MultipleLinearRegressionof‘Penetration[mm]’

MultipleLinearRegressionof‘Penetration[mm]’

Projectionoftheresponsesurface

MultipleLinearRegressionof‘Penetration[mm]’

LSL

Penetrationvs.twooftheimportantx’s

AlltogetherOptimalsettingswhenGapis….

LSL

11,5

22,5

3

Pred

For

mul

a Fl

ange

Toe

R

adiu

s [m

m]

0,959241

LSL12345

Pred

For

mul

a W

eb T

oe R

adiu

s [m

m]

2,116754

1LSL

3456

Pred

For

mul

a Pe

netra

tion

[mm

]

3,26743

4LSL

Target5,5

USL6,5

Pred

For

mul

a Be

ad T

hick

ness

[m

m]

5,455627

-1,5-1

LSLTarget

USL1

1,52

2,5

Pred

For

mul

a D

evia

tion

from

Fl

at [m

m]

-0,05867

00,

250,

751

Des

irabi

lity

0,439824

0

0,5 1

1,5 2

0Gap [mm]

35 40 45 50 55

56,777778Web Plate tovertical [°]

-10 -5 0 5 10 15 20

-7,759601Up (+) / Down(-) slope [°]

80 90 100

110

120

120Gun angel aboveunwelded seam [°]

35 40 45 50 55

55,444444Gun angle toWeb Plate [°]

0

0,25 0,

5

0,75 1

Desirability

WithaGapof0mmDesirabilitybelowshowthatUp(+)/Down(-)slopeisthemostcriticalfactortokeepright.Togetherwithastrongpushwelding.

PushdownhillwhenGapissmall

ally’s

vs.allx’s

LSL

11,5

22,5

3

Pred

For

mul

a Fl

ange

Toe

R

adiu

s [m

m]

1,564007

LSL12345

Pred

For

mul

a W

eb T

oe R

adiu

s [m

m]

3,110296

1LSL

3456

Pred

For

mul

a Pe

netra

tion

[mm

]

4,391701

4LSL

Target5,5

USL6,5

Pred

For

mul

a Be

ad T

hick

ness

[m

m]

5,381489

-1,5-1

LSLTarget

USL1

1,52

2,5

Pred

For

mul

a D

evia

tion

from

Fl

at [m

m]

-0,03529

00,

250,

751

Des

irabi

lity

0,629058

0

0,5 1

1,5 2

1Gap [mm]

35 40 45 50 55

56,777778Web Plate tovertical [°]

-10 -5 0 5 10 15 20

-0,002457Up (+) / Down(-) slope [°]

80 90 100

110

120

119,98446Gun angel aboveunwelded seam [°]

35 40 45 50 55

55,444444Gun angle toWeb Plate [°]

0

0,25 0,

5

0,75 1

Desirability

WithaGapof1mmStillsamefactorbutnotsocriticaltogetherwithastrongpushwelding.

PushhorizontalwhenGapismedium

ally’s

vs.allx’s

LSL

11,5

22,5

3

Pred

For

mul

a Fl

ange

Toe

R

adiu

s [m

m]

2,42022

LSL12345

Pred

For

mul

a W

eb T

oe R

adiu

s [m

m]

3,053831

1LSL

3456

Pred

For

mul

a Pe

netra

tion

[mm

]

5,43873

4LSL

Target5,5

USL6,5

Pred

For

mul

a Be

ad T

hick

ness

[m

m]

4,913004

-1,5-1

LSLTarget

USL1

1,52

2,5

Pred

For

mul

a D

evia

tion

from

Fl

at [m

m]

0,038164

00,

250,

751

Des

irabi

lity

0,815613

0

0,5 1

1,5 2

2Gap [mm]

35 40 45 50 55

55Web Plate tovertical [°]

-10 -5 0 5 10 15 20

8,4084818Up (+) / Down(-) slope [°]

80 90 100

110

120

86,192858Gun angel aboveunwelded seam [°]

35 40 45 50 55

55Gun angle toWeb Plate [°]

0

0,25 0,

5

0,75 1

Desirability

WithaGapof2mmTheprocesswindowchangetopullinguphillinsteadinordertokeepifthebeadgeometryrequirementsontarget.

PulluphillwhenGapisbig!

ally’s

vs.allx’s

SimulationofprocesscapabilityusingMonteCarlo

Addingrandomnoisetothepredictors…

Outgoingcapabilitycanbepredicted

Modelverification

Modelsaregoodcomparedwithreferencesampleforpenetration,a-dimandDeviationfromflat(lowerthree)

AndslightlyofffortheToeradii,whichneedsomefurtheranalysis.

Conclusions

• ControlfactorsforProductivityandQualitycanpotentiallybeseparated:• Firstoptimizeprocessspeedandmetallurgy• ThenutilizethedegreesoffreedomintheRobotManipulatorsinordertosetweldinggeometryanglestoshapeBeadandcompensateforprocessvariations,thatis,varyingGAP,improvingrobustness

• Processwindowexiststomeetallrequirements• Atleastfora5mmFilletWeldweldedwithWPS15

• CostislimitedusingefficientDOEmethodologiesexploringtheexperimentalset-up

• 13+3samplesinordertoevaluate6controlfactors

• PITFALLS:Baddata!• Measurementnoise<30%oftolerances

• Ifthetoleranceofatoeradiusis0,3mmtheprecisionofthemeasuresneedstobe<0,1mm…• Fixedduringoff-lineset-upwithrepeatedmeasurements• On-line….hmmm

Futurework

• Improverobustnesstoincomingvariationwithaself-adjustingweldingsystemusing:

• in-linemonitoringofincomingvariationinx’s• trailingmonitoringofoutgoinggeometry(laserscanning)• defineprocesswindowsusingpredictivemodelsfromoff-lineexperiments• utilising Machinelearningalgorithms–orinlineadjustment• 6-axisrobots

• PotentialreductionofQualityassessmentneeds• Increasedknowledgeofthecorrelationbetweenouterandinnerweldgeometrywouldreducetheneedtomonitorhard-to-measureparameters(…penetration…)

• Whymonitorthingsthatcanbepredicted…let’sconcentrateonthedifficultones…

? & !

References

• Suh,NamPyo.(2001).Axiomaticdesign,advancesandapplications.OxfordUniversityPress,NewYork.ISBN0-19-513466-4

• JonesB.,Nachtsheim C.J.,AClassofThree-LevelDesignsforDefinitiveScreeninginthePresenceofSecond-OrderEffects,JournalofQualityTechnology,Vol.43,No.1,January2011